Serially measured blood biomarkers collected in the context of usual care, predict clinical outcome in a real-life heart failure population

Abstract

BACKGROUND: Studies on multiple serially measured blood biomarkers and adverse outcomes in patients with heart failure (HF) are scarce and restricted to research settings. Prognostic estimates based on serial measurements could provide a scientifically substantiated, uniform approach to risk stratification and timing of treatment. We use an exploratory and hypothesis-generating approach to assess the predictive ability of serially measured biomarkers, commonly collected during usual care, for adverse events in a real-world population of ambulant patients with HF with reduced ejection fraction (HFrEF) and HF with mildly reduced ejection fraction (HFmrEF). METHODS: We included ambulatory HFrEF-HFmrEF patients who attended an outpatient hospital visit between 2017 and 2022. Data on blood biomarkers, clinical characteristics and clinical outcome were extracted from electronic health records. Joint modelling was applied to investigate associations between time-varying blood biomarkers and the composite endpoint of mortality, LVAD implantation and heart transplant. RESULTS: We included 1353 patients; 66.8% men; median(P25, P75) age 62(51, 71) years. During a median follow-up of 3.30(1.62, 4.65) years, 387(28.6%) experienced the endpoint. Temporal trajectories of 30 blood biomarkers were significantly associated with the endpoint. After correcting for clinical characteristics, associations persisted in multiple-biomarker models for serially measured NT-proBNP, hs-TnT, and CRP, as well as liver biomarkers, kidney biomarkers, and blood count parameters. Serial measurements increased model performance compared to baseline measurements, with AUCs up to 0.83 for multiple-biomarker models. CONCLUSION: Real-life, serially measured laboratory data predict adverse events in an ambulatory HFrEF-HFmrEF population, with good internal model performance. Thus, making use of laboratory values already present in electronic medical records, could inform risk stratification without any extra effort.